Random Utility Models For Social Choice
Abstract: Social choice studies ordinal preference aggregation with wide applications ranging from high-stakes political elections to low-stakes movie rating websites. In many cases, we want to design a social choice mechanism that reveals the ground truth via MLE inference. Despite its importance, this objective has been largely hindered by the lack of natural probabilistic models and efficient inference algorithms.
In this talk, I will focus on a wide class of probabilistic models called Random Utility Models (RUMs), whose MLE inference was previously believed intractable in general. I will introduce a fast MC-EM algorithm for a very general and natural subclass of RUMs, and discuss its applications and impact on designing better social choice mechanisms. Extension of this algorithm also provides a computational basis for improving models in many applications in economics as well as machine learning, especially learning to rank.
Based on joint work with Hossein Azari Soufiani and David C. Parkes.
Short bio: Lirong Xia is a CRCS fellow and NSF CIFellow at the Center for Research on Computation and Society (CRCS) at Harvard University. He received his Ph.D. in Computer Science in 2011 and M.A. in Economics in 2010, both from Duke University. His research focuses on the intersection of computer science and microeconomics, in particular computational social choice, game theory, mechanism design, and prediction markets.